With the increasing computing power of edge devices, Federated Learning (FL)
emerges to enable model training without privacy concerns. The majority of
existing studies assume the data are fully labeled on the client side. In
practice, however, the amount of labeled data is often limited. Recently,
federated semi-supervised learning (FSSL) is explored as a way to effectively
utilize unlabeled data during training. In this work, we propose ProtoFSSL, a
novel FSSL approach based on prototypical networks. In ProtoFSSL, clients share
knowledge with each other via lightweight prototypes, which prevents the local
models from diverging. For computing loss on unlabeled data, each client
creates accurate pseudo-labels based on shared prototypes. Jointly with labeled
data, the pseudo-labels provide training signals for local prototypes. Compared
to a FSSL approach based on weight sharing, the prototype-based inter-client
knowledge sharing significantly reduces both communication and computation
costs, enabling more frequent knowledge sharing between more clients for better
accuracy. In multiple datasets, ProtoFSSL results in higher accuracy compared
to the recent FSSL methods with and without knowledge sharing, such as
FixMatch, FedRGD, and FedMatch. On SVHN dataset, ProtoFSSL performs comparably
to fully supervised FL methods.

ProtoFSSL 是一种新颖的基于原型网络的半监督 FL 方法，它通过轻量级原型实现客户端之间的知识共享，减少了通信和计算成本，并在多个数据集上实现了更高的准确性。